Toward the next frontier of embodied AI

1. INTRODUCTION

Recently, embodied artificial intelligence has emerged as a transformative paradigm in artificial intelligence research, marking a significant shift toward systems that tightly couple perception, action, and interaction within physical environments^[1]. Progress in this field is fueled by interdisciplinary insights and state-of-the-art technologies, enabling the development of adaptive, autonomous agents that can navigate and learn from complex, real-world scenarios^[2]. This evolution signals a new chapter in artificial intelligence, wherein machines transcend passive observation to actively engage with their environments, exhibiting levels of situational understanding that increasingly resemble human cognitive processes^[3].

Taking autonomous driving as an example, embodied artificial intelligence enables vehicles not only to perceive their surroundings through advanced sensors such as LiDAR, Radar, and cameras, but also to dynamically adapt to evolving traffic conditions, interact safely with pedestrians, and respond effectively to unforeseen events^[4]. These intelligent agents leverage contextual understanding to make real-time decisions that prioritize both safety and operational efficiency^[5]. Similarly, service robots equipped with embodied intelligence can skillfully navigate domestic and public environments, comprehend natural language instructions, and engage in complex interactions with both objects and humans^[6]. Such advancements mark a significant step toward the meaningful integration of artificial intelligence into real-world environments.

This Special Issue aims to demonstrate recent advances in the rapidly evolving field of embodied artificial intelligence. We received eight submissions from researchers across the globe, reflecting the growing interest and momentum in this area. Following a rigorous peer-review process and valuable feedback from expert reviewers, four articles were selected for publication. These contributions primarily address key challenges in perception and localization, presenting novel insights and methodologies that advance the capabilities of intelligent embodied systems.

2. CONTRIBUTED ARTICLES

Among the four accepted articles, three deep learning-based studies focus on leveraging single-modal neural networks to perform salient object detection^[7], facial expression recognition^[8], and infrared object detection^[9], respectively. Although targeting different application domains, these studies share a unified vision: enabling machines to perceive, interpret, and respond to complex visual environments in real time. Each work introduces specialized network architectures designed to enhance perception accuracy while maintaining computational efficiency, reflecting a collective push toward deployable artificial intelligence in resource-constrained or dynamic conditions. A notable commonality among these studies lies in their emphasis on multi-scale feature learning. Whether through attention mechanisms, convolution-transformer fusion, or scale-adaptive modules, all three approaches integrate hierarchical visual cues to capture both global semantic context and fine-grained structural detail. This multi-scale strategy enables a crucial balance between semantic comprehension and spatial precision, an essential attribute for embodied artificial intelligence systems operating in unstructured or variable environments. Furthermore, each method demonstrates strong empirical performance on public benchmark datasets, underscoring the robustness and generalizability of the proposed designs.

Despite these shared foundations, each study makes distinct and complementary contributions. The study^[7] focuses on lightweight salient object detection, introducing scale-adaptive feature extraction and multi-scale feature aggregation modules to achieve an optimal trade-off between efficiency and accuracy. The study^[8] addresses facial expression recognition toward embodied artificial intelligence, proposing a multi-scale attention and convolution-transformer fusion network to enhance emotion-aware human-robot interaction^[10]. Finally, the study^[9] targets infrared object detection under adverse weather conditions, combining the MobileNetV3-YOLOv4 architecture with an image enhancement generative adversarial network to ensure high-precision detection on low-power edge devices.

Taken together, these studies collectively advance the frontier of efficient and adaptive visual perception. By addressing complementary aspects of perception, ranging from the semantic understanding of human emotions to the structural saliency of objects and the multi-modal robustness required under low-visibility conditions, they demonstrate a coherent progression toward unified, context-aware perception systems. Such efforts not only contribute to academic exploration but also carry significant practical implications for real-world applications, from intelligent vehicles and service robots to next-generation embodied agents capable of understanding and interacting with their surroundings in a human-like manner.

In addition, the remaining study^[11] explores the parallel implementation of a real-time visual simultaneous localization and mapping system through heterogeneous parallel computing. Although it does not rely on deep learning, the study shares several conceptual commonalities with the aforementioned works. First, it emphasizes computational efficiency and real-time performance, an objective aligned with the other contributions that enhance learning and perception efficiency through architectural optimization or multi-modal fusion. Second, similar to the deep learning-based studies, it contributes to the advancement of embodied intelligence and autonomous systems, where robust perception and low-latency computation are critical for deployment in complex real-world environments. Third, all four studies integrate vision with high-performance computing strategies to manage large-scale visual data: the deep learning approaches do so through large vision models and multi-modal learning, while this work adopts hardware-level parallelization. Overall, this study complements the learning-based approaches by addressing computational bottlenecks from a systems and hardware perspective, underscoring the importance of real-time, scalable visual processing for embodied artificial intelligence.

3. EXISTING CHALLENGES AND FUTURE RESEARCH TRENDS

Despite the remarkable progress in embodied artificial intelligence, several key challenges remain unresolved. First, most existing frameworks exhibit limited generalization and adaptability when deployed in diverse real-world environments. While deep learning-based methods have demonstrated impressive capabilities, they often depend heavily on large-scale annotated datasets and struggle with continual learning or domain transfer. This reliance frequently leads to catastrophic forgetting during long-term autonomous operation. Second, current multi-modal perception systems lack efficient mechanisms for cross-modal understanding and fusion, especially in dynamic scenarios where heterogeneous sensor data present misaligned spatio-temporal characteristics. Third, intelligent systems face significant constraints related to computational and energy efficiency. The widespread dependence on large-scale vision-language models and complex neural architectures hampers their real-time applicability on resource-constrained platforms, such as mobile robots, aerial drones, and intelligent vehicles. Finally, there remains a gap between algorithmic innovation and hardware-level optimization. Most research emphasizes model design while neglecting the computational parallelization, scheduling, and optimization strategies required for efficient deployment on heterogeneous devices.

Future research in embodied artificial intelligence is expected to evolve along several promising directions. First, lifelong and continuous learning will become a central objective, enabling embodied agents to incrementally acquire new skills and adapt to changing environments without retraining from scratch^[12]. Second, multi-modal and cross-domain integration will deepen, aiming to build unified representations that effectively combine vision, language, and sensory cues for robust reasoning and decision-making^[13-15]. Third, the development of large-scale foundation models specifically tailored for embodied perception, particularly those inspired by the human dual-stream visual processing hypothesis, will catalyze progress toward generalizable understanding and interaction across diverse tasks and contexts^[16]. Fourth, the design of lightweight and energy-efficient models will be essential for deploying complex visual-language architectures on embedded platforms, necessitating advances in compression, pruning, and knowledge distillation techniques^[17]. Fifth, hardware-software co-optimization is expected to receive growing attention, with the integration of parallel computing strategies, such as Compute Unified Device Architecture (CUDA)-based heterogeneous acceleration and graphics processing unit (GPU)/tensor processing unit (TPU) adaptation, playing a vital role in addressing the computational bottlenecks that currently constrain real-time embodied intelligence^[11]. Finally, future embodied artificial intelligence systems must prioritize interpretability and safety to ensure transparent decision-making and foster trustworthy interactions between intelligent agents and humans in complex, unstructured environments.

4. CONCLUSION

In summary, this editorial underscored the pressing need for advancing embodied artificial intelligence, highlighted the key contributions of the accepted articles in this special issue, and briefly discussed both the current challenges and promising future research directions in the field.